The Economics of App Review∗

Ming Gao† Travis Ng‡

February 17, 2016

Abstract

A platform for apps, such as Apple App Store, facilitates the interaction between

app users and app developers. For political, moral, and various other reasons, some

apps may cause more trouble to the platform than others even if the users may not

value them differently. In a two-sided market model with both “well-behaved” and

“trouble-making” developers, we show that, whereas an app review is useful for

screening trouble makers, the platform faces a trade-off between cost saving (from

less trouble) and a loss of the attractiveness to the users (from reduced app variety).

The platform’s incentive to use app review depends on the additional cost it incurs

from each “trouble maker”, and on their proportion among all developers. Given the

additional cost incurred, app review benefits the platform if and only if the propor-

tion of trouble makers is sufficiently small, such that after screening them the platform

remains fairly attractive to users. Given their proportion, however, app review helps

if and only if the “trouble makers” create enough “trouble”, such that the total cost

saved from app review is not negligible. Our welfare analysis through simulations

cautions regulators that the platform’s incentive to use app review may be aligned or

misaligned with the interest of social welfare, depending on the distributions of app

users and developers, the strength of network effects and other parameters.

Key Words: app review, Apple, Android, two-sided market, non-price discrimi-nation

JEL Classification: D42, L11, L12

∗Preliminary and incomplete. We thank seminar participants at Shanghai Jiao Tong University andShanghai University of Finance and Economics. Ming Gao acknowledges the financial assistance of Project20151080389 supported by the Tsinghua University Initiative Scientific Research Program and Project71203113 supported by the National Natural Science Foundation of China.†Corresponding author: School of Economics and Management, Tsinghua University, Email:

gaom@sem.tsinghua.edu.cn.‡Department of Economics, The Chinese University of Hong Kong.

1 Introduction

According to ABI Research the smartphone app market was worth US$27 billion in 2013.1

App distribution platforms for smartphones, such as App Store and Google Play, are ex-amples of two-sided platforms where smartphone users and app developers interact witheach other, and the benefits each side derives depend directly on the number of agentson the other side. That is, other things being equal, smartphone users usually prefer plat-forms with more apps, whereas app developers prefer platforms with more users. Rochetand Tirole (2003 and 2006) and Armstrong (2006) provide some of the canonical modelsfor two-sided markets. Kouris (2011) models specific characteristics of app distributionplatforms and studies optimal pricing strategies of such platforms. His model, however,does not analyze the effect of any non-price strategy on market outcomes.

The app review process is one non-price strategy that Apple uses. While App Storeand Google Play are the two dominant app distribution platforms, they used to have amajor difference, i.e. there has long been an ex ante review process on App Store, butnot on Google Play until 2015.2 Why the difference? To answer this question, this paperbuilds a monopoly model of two-sided market and analyzes the effect of Apple’s appreview process on app users, developers, App Store and social welfare.3

In some two-sided markets, platforms use quality as a criterion to exclude some agentson one or both sides from entering the platforms. For instance, night clubs are likely toforbid guys and girls looking lousy from entering. Hagiu (2009) provides a model ofusing quality as an exclusion criterion and investigates the conditions for the impositionof such an exclusion system.

Is quality the main selection criterion for the App Store’s review process? We argue that itis unlikely for three reasons: First, the variety of existing apps just seems too enormousto allow for a concensus on the objective measures for their quality. Second, the moreinnovative (and hence different from other apps) a new app is, the more difficult it seemsto gauge its quality.4 Third, many examples suggest that apps can be rejected for reasonsthat are apparently unrelated to quality (e.g. Google Map was excluded for quite some

1Source: https://www.abiresearch.com/press/the-mobile-app-market-will-be-worth-27-billion-in-2Google Play did not introduce a review process for apps until early 2015. Source: http://www.

theverge.com/2015/3/17/8231125/android-apps-now-reviewed-by-google3Apple describes its App Store review guidelines here: https://developer.apple.com/appstore/

guidelines.html4We are referring to “user quality” rather than what Apple might define as quality. Consider gambling

apps, for instance, which are likely valued highly by just a small proportion of all app users. It is probablyunfair to say that a new gambling app must have low quality just because most users might not evenconsider using it. But even if most gamblers would use this new app, is it fair then to say that it must havehigh quality? What about an app with a completely new function that cannot even be categorized? Howshould its quality be measured?

1

https://www.abiresearch.com/press/the-mobile-app-market-will-be-worth-27-billion-in-http://www.theverge.com/2015/3/17/8231125/android-apps-now-reviewed-by-googlehttp://www.theverge.com/2015/3/17/8231125/android-apps-now-reviewed-by-googlehttps://developer.apple.com/appstore/guidelines.htmlhttps://developer.apple.com/appstore/guidelines.html

time by Apple).5

For example, some apps are rejected because they duplicate existing apps, their userinterface designs do not meet Apple’s requirements (which do not necessarily mean theyare of lower quality), they contain objectionable content (which some users may valuehighly), or they mention other supporting platforms. Apple uses its own set of implicitstandards to review apps which are not necessarily related to quality.

We construct a model different from Haigu (2009) to investigate how an app reviewprocess affects smartphone users, app developers and the platform. We derive conditionsfor the incentive of a monopoly app distribution platform to review apps, and study thewelfare implications.

We assume there are two types of app developers who are indistinguishable to theplatform before they are reviewed. For political, moral, and/or various other reasons, onetype causes more trouble to the platform and is more costly to serve than the others. Forinstance, being associated with certain politically sensitive or controversial topics maybe bad for the platform’s public image, and it may therefore consider it troublesome ifdevelopers show them in their apps. Once an app is submitted for review, however, thetrue type of its developer will be revealed to the platform. If the platform pleases, it canuse the review process to discourage participation by developers that are more costly toserve. For instance, the platform can repeatedly reject such developers’ apps or simplyprolong the review process such that some of them will eventually give up.

In other words, app review can be used to destroy value of developers who are morecostly to the platform. But such developers are not necessarily less valuable to app users.We focus on the situation where users have a preference for variety that is reflected bythe total number of developers on the platform. Therefore, when screening more costlydevelopers through app review, the platform is faced with a trade-off between the costsaving and the loss of its overall attractiveness to app users.

We develop a simple model à la Armstrong (2006) and find that the platform’s incen-tive to review apps depends on the additional cost it incurs due to each “trouble-making”developer, and on their proportion among all developers. Given the additional cost in-curred, app review benefits the platform if and only if the proportion of trouble makersis sufficiently small, such that after screening them the platform remains fairly attractiveto users. Given their proportion, however, app review helps if and only if the “trouble

5For some useful comments on App Store’s review process, read: http://en.wikipedia.org/wiki/Approval_of_iOS_apps and http://www.tuaw.com/2008/08/07/thoughts-on-the-iphone-app-store-review-process/. The first website pro-vides some examples of rejected apps. http://venturebeat.com/2013/02/08/9-surprising-reasons-mobile-apps-get-rejected-from-the-apple-app-store/ provides 9 sur-prising reasons for an app to get rejected from App Store.

2

http://en.wikipedia.org/wiki/Approval_of_iOS_appshttp://en.wikipedia.org/wiki/Approval_of_iOS_appshttp://www.tuaw.com/2008/08/07/thoughts-on-the-iphone-app-store-review-process/http://www.tuaw.com/2008/08/07/thoughts-on-the-iphone-app-store-review-process/http://venturebeat.com/2013/02/08/9-surprising-reasons-mobile-apps-get-rejected-from-the-apple-app-store/http://venturebeat.com/2013/02/08/9-surprising-reasons-mobile-apps-get-rejected-from-the-apple-app-store/

makers” create enough “trouble”, such that the total cost saved from app review is notnegligible.

Our welfare analysis through half a million simulations cautions regulators that theplatform’s incentive to use app review may be aligned or misaligned with the interestof social welfare, depending on the distributions of app users and developers and otherparameters, including the strength of network effects.

It remains a black box to outsiders as to why exactly an app platform like Apple im-poses an ex ante review process. We are simply suggesting one arguably straight-forwardway to look at it. Is it a complete view? We do not think so. But to the best of our knowl-edge, there is yet any work that completely decodes the platform’s true incentives for appreview. This may be the reason why even experienced app developers would get rejectedby App Store from time to time. Even though Apple has an official website for app re-view rules and statistics of recent review outcomes, it does not reveal clear reasons for themajority of rejections.6 It is plausible that some of the apps that got rejected are highlyvaluable to some (if not most) users (such as Google Map, Bitcoin and Google Voice).Therefore, app quality to users, though stated officially by Apple as a criterion, cannot beregarded as their sole criterion for screening. Trouble-making appears a reasonable usualsuspect, where troubles to Apple by an app can be broadly defined as including every-thing from political position, to moral (such as apps with indecent content), to marketpower, etc. Appendix B gives more details.

Although we construct a monopoly model, we certainly note that Google competeswith Apple in app distribution. We do not think that the competition among them isirrelevant. However it remains a challenge to us to properly incorporate the followingthree prominent market features into a duopoly or oligopoly model: a) multihoming byboth sides of the market (such as the fact that many users carry an iphone for personaluse and an Android phone for business, and that seemingly few developers work exclu-sively on IOS but not Android, or vice versa); b) the difference in the platforms’ mainrevenue sources (Google’s is mainly Ad revenue, as indicated in the Appendix, but Ap-ple’s is very different); c) Apple being highly vertically integrated in terms of hardwareand software, whereas Google heavily concentrating on smartphone OS than hardware.

6It may seem that Apple has a transparent and self-explanatory app review standard, with a long listof guidelines clearly written in black and white. However in January 2016, Apple announced the verytop one among 10 most common reasons for rejections from App Store, accounting for 44% of all, was“Other reasons”, without revealing what exactly those other reasons were. (https://developer.apple.com/app-store/review/rejections/) And the second most common reason, taking up 16%, was “Moreinformation needed.” These two obscure categories together account for 60% of all rejections. Some clearreasons for app rejections that Apple did reveal included “bugs” (12%), “complex or less than very good”(6%), “irrelevant content and functionality” (4%), and “crashes” (3%).

3

https://developer.apple.com/app-store/review/rejections/https://developer.apple.com/app-store/review/rejections/

One additional benefit of our monopoly model, though, is that it can be contrasted easilywith Armstrong’s (2006) monopoly model and many other models in the literature thatare based on his.

2 The model

Basic set-up

There is a monopoly app distribution platform which facilitates interaction betweentwo groups of agents, where each group is represented by a continuum. Group-1 agentsare app users and group-2 agents are app developers. Denote n1 and n2 the numbers ofagents eventually joining the platform from group 1 and 2, respectively. The platformcharges a fixed price p1 and p2 to each agent in group 1 and 2, respectively, for access tothe platform.

App users have a preference for app variety. The utility that an app user derives fromparticipation on the platform is

u1 = (α − γ)n2 − p1, (1)

where α(> 0) represents the network benefit she enjoys from interacting with each de-veloper, and γ(≥ 0) represents the charge by each developer. We treat γ as an exogenousparameter.7

Without app review, an app developer’s utility is given by

u2 = γ(1 − β)n1 − p2, (2)

where β ∈ [0, 1] represents the commission that the platform takes from each developer’srevenue.

Following Armstrong (2006), we assume that the number of app users depends solelyon their utilities, denoted

n1 = φ1(u1).

where φ1 is strictly increasing.8

Two developer types

7We choose not to model any developer’s optimization process or the competitive situation amongdevelopers, but focus on the platform’s incentives. In real life, most apps are provided for free and thereforeone can set γ equal to 0. We however also allow for the possibility of a positive exogenous γ.

8We provide a specific functional form of φ1 in section 5.

4

The group of developers consists of two types. Type 1 are “well-behaved” developersand their proportion is λ ∈ [0, 1]; the remaining (1 − λ) proportion are type-2 developerswho are “trouble makers.” Denote n2k the number of type-k developers on the platform(k = 1, 2), and assume that, when a developer derives utility u2 from participation in theplatform, the numbers of different developers are given by

Type-1 developers: n21 = λ · φ2(u2),Type-2 developers: n22 = (1 − λ) · φ2(u2),

where φ2 is strictly increasing. The total number of all developers on the platform istherefore

n2 = n21 + n22.

The two types are equally valuable in the eyes of app users, and they only differ interms of the cost that the platform incurs to serve them. In particular, it costs f21 to serveeach type-1 developer and f22 to serve each type 2. Denote

∆ f ≡ f22 − f21.

Suppose it costs the platform f1 to serve each app user. Without loss of generality, assumef1, f21, f22,∆ f ≥ 0.

Introducing app review

We model the app review process as designed to discourage trouble-making develop-ers’ participation. In particular, app review results in a reduction in the utility that type-2developers derive from the platform, denoted by R ∈ [0, R̄], whereas app review does notaffect the utility of any type-1 developers.9 The upper bound R̄(> 0) on R simply indicatesthat the platform can only reduce type-2 developers’ utility within a reasonable range,but not infinitely.

Therefore, when each type-1 developer derives u2 from the platform, the utility thateach type-2 developer derives will be u2 − R as long as the platform uses app review. Thenumber of type-2 developers is summarized as follows for a given utility u2 derived bytype-1 developers:

without app review: n22 = (1 − λ) · φ2(u2),with app review: n22 = (1 − λ) · φ2(u2 − R).

9In the extension in section 4 we relax this assumption and allow for the possibility that app reviewmay imperfectly target type-2 developers.

5

The platform’s profit is given by

π = (p1 − f1)n1 + (p2 − f21)n21 + (p2 − f22)n22 + βγn1(n21 + n22), (3)

3 When to Review Apps?

We now focus on the impact of app review on the platform’s profitability. To facilitatea clear analysis, we consider the utilities that the platform provides for different sides,u1 and u2, as its choice variables, and treat the utility destroyed by app review to type-2developers, R, as an exogenous parameter. This approach allows us to show how theplatform’s maximized profit changes at different R. Whether or not to use app reviewremains the platform’s choice to make. When it does, R will be deducted from type-2developers’ utility, and when it does not, there will be no deduction.

Rewrite expression (3) as a function of u1, u2 and R as follows:

π(u1, u2,R) ≡ (p1 − f1)n1 + (p2 − f21)n21 + (p2 − f21 − ∆ f )n22 + βγn1(n21 + n22), (4)

where

n1 = φ1(u1);

n21 = λφ2(u2);

n22 = (1 − λ)φ2(u2 − R);n2 = n21 + n22;

p1 = (α − γ)(n21 + n22) − u1;p2 = γ(1 − β)n1 − u2.

The first-order condition of (4) with respect to u1 is

(αn2 − u1 − f1)︸ ︷︷ ︸economic profit from each user

· dn1du1︸︷︷︸

rise in n1

= n1︸︷︷︸loss in revenue

(5)

where dn1du1 = φ′1(u1).

6

The first-order condition of (4) with respect to u2 is

(αn1 − u2 − f22)︸ ︷︷ ︸economic profit from

each type-2 developer

· ∂n2∂u2︸︷︷︸

rise in # of

developers

= n2︸︷︷︸loss in

revenue

− ∆ f︸︷︷︸cost saved by

each type-1

· dn21du2︸︷︷︸

rise in #

of type-1

(6)

where ∂n2∂u2

= λφ′2(u2) + (1 − λ)φ′2(u2 − R) anddn21du2

= λφ′2(u2).Each developer creates a total value of αn1 for the platform, and therefore the economic

profit that a type-2 developer generates is αn1 − u2 − f22. When the platform raises u2, atotal of ∂n2

∂u2new developers join the platform, and the left-hand side of (6) represents the

total increase in economic profit from these new developers, if they were all of type 2. Onthe right-hand side, the gross loss in revenue due to the additional utility offered to alldevelopers equals their number n2. But as a fraction of dn21du2 new developers are of type 1who each save the platform ∆ f in cost, the total cost saving from them is equal to ∆ f · dn21du2 ,which needs to be deducted from the gross loss in revenue.

Given some R ∈ [0, R̄], denote the u1 that satisfies (5) and u2 that satisfies (6) as thefollowing

(u∗1(R), u∗2(R)) ≡ arg max(u1,u2)

π(u1, u2,R)

and denote the maximized profit as a function of R alone as follows

Π(R) ≡ π(u∗1(R), u∗2(R),R) (7)

Assume that the profit function (4) is well-behaved such that u∗1(R) and u∗2(R) are dif-

ferentiable for R ∈ [0, R̄]. This implies that Π(R) is also differentiable on this interval. Wehave the following useful property of Π(R).

Lemma 1 Π(R) is U-shaped in R. That is, Π′′(R) > 0 if Π′(R) = 0.

(All omitted proofs are provided in the Appendix.) Lemma 1 implies that the maxi-mized profit as a function of R has a single trough but no peak. Because the maximizedprofit without app review is equal to Π(0), we can use Π′(0) and Lemma 1 to determinethe platform’s incentive to use app review. For R ∈ [0, R̄], apply an envelope argument for(7) and we must have

Π′(R) =d

dRπ(u∗1(R), u

∗2(R),R) =

∂

∂Rπ(u∗1(R), u

∗2(R),R) (8)

7

Therefore we can find Π′(0) by evaluating ∂π∂R at the optimal utilities without app re-

view.

Proposition 1 (Optimal Pricing without App Review) Without app review, the optimal pricesfor two sides, (p01, p

02), are given by p

01 − f1 + γn2 =

n1n′1

p02 − ( f22 − λ∆ f ) + [α − γ(1 − β)]n1 =n2n′2

(9)

and the optimal utilities that the platform provides to two sides, (u01, u02), are given by φ1(u01) = φ′1(u01)(αφ2(u02) − u01 − f1)φ2(u02) = φ′2(u02)(αφ1(u01) − u02 − f22 + λ∆ f ) (10)

Without review, the platform has no way to treat the two types of developers differ-ently, and therefore the utility it provides to them is determined by the average cost thatthe platform incurs due to each developer, ( f22 − λ∆ f ).

The partial derivative of (4) with respect to R gives

∂

∂Rπ(u1, u2,R) =

∂n22∂R

(αn1 − u2 − f22) (11)

where ∂n22∂R = −(1−λ)φ′2(u2−R). Intuitively, raising R has the exact opposite effect on type-2

developers as raising u2 does. The negative impact on profitability is the product betweenthe decrease in the number of type-2s and the economic profit they each create.

Without app review, n2 = φ2(u02) and∂n2∂u2

= φ′2(u02). At optimality, by the first-order

condition (6) we know, the platform will choose the optimal price (and utility) for devel-opers to balance the economic profit from type-2 developers and the net loss it incurs perdeveloper when raising u2, that is,

(αn1 − u02 − f22) =n2 − ∆ f · dn21du2

∂n2∂u2

Substitute in (11) and by (8) we have

Π′(0) = (1 − λ)[λ∆ fφ′2(u02) − φ2(u02)].

Therefore by Lemma 1, we have the following result.

8

Proposition 2 (Incentive to Review Apps) The platform has an incentive to introduce app re-view if and only if

λ∆ f ≥φ2(u02)

φ′2(u02)

(12)

where u02 is the utility that each developer obtains from the platform without app review, as givenby (10).

Intuition: Compared to a type-2 developer, each type-1 developer saves the platform∆ f in cost, whilst creating the same network benefits for app users on the opposite side.Because the proportion of type-1 developers is λ, λ ·∆ f represents the total cost saved dueto existing type-1 developers on the platform, compared to the “worst” situation whereall developers are of type 2. Proposition 2 implies that app review is profitable if andonly if the total cost saving from existing good developers is significant enough. In otherwords, app review works in favor of the platform if and only if the problem created bytrouble-making developers is mild.

For a given level of ∆ f > 0, when the proportion of good (type-1) developers, λ, issufficiently large, it is profitable for the platform to introduce app review to discourageparticipation by type-2 developers. The gain comes from the cost saving due to sometype-2 developers leaving the platform as a result of app review. The loss results fromthe reduced attractiveness of the platform to app users due to fewer developers overall.Introducing app review only makes sense when the proportion of type-2 developers issufficiently small such that the platform will still remain fairly attractive to users evenafter losing some of them.

Similarly, for a given proportion λ > 0 of good developers, app review is profitablewhen ∆ f is big enough, i.e. if type-2 developers create enough trouble for the platform tojustify its interference.

App review is intended to reduce the presence of trouble-making developers on theplatform. When the additional cost ∆ f they bring is too small, or when their proportion(1− λ) is too large, the problem caused by them is either negligible or too severe. In eithercase, the cost saved due to existing type-1 developers will be too small, which rendersapp review ineffective.

Positive threshold

The positive threshold φ2(u02)

φ′2(u02)

in (12) may explain why Google and Apple adopted dif-ferent app review strategies. Google did not use any app review measures until quiterecently even though there were most likely trouble-making developers on it platform all

9

along. It is possible that Google has a smaller λ, perhaps due to the variety of developersattracted to its more open operating system.10

A close look at Lemma 1 reveals that, if it is possible to induce a negative R for type-2developers, perhaps by some kind of “reverse review” (as compared to the normal appreview that results in a positive R), the platform should do so whenever λ∆ f < φ2(u

02)

φ′2(u02)

.Intuitively, this happens either when there are too many bad developers (λ small), or theadditional cost ∆ f they cause is too small. Given that the platform has already found theoptimal u∗2(R) for good type-1 developers, it would make sense for the platform to use thenegative R to further increase the utility offered to bad developers, in order to attract moreapp users on the opposite side. Note this only makes sense because, besides u∗2(R), theplatform has no other way to further increase the utility offered to the good developers.If it had other tools, it would certainly rather use them to raise good developers’ utilityfirst. We do not really observe such “reverse review” in real life, maybe because gooddevelopers already pass the app review 100% of the time, and it is impossible to providethe bad developers an even better chance at passing it.

Also note that when λ∆ f = φ2(u02)

φ′2(u02)

, by (9) and (10) we have

p02 = f22 − [α − γ(1 − β)]n1

Because [α − γ(1 − β)]n1 exactly represents the total network benefits that a developercreates for the platform, f22 − [α − γ(1 − β)]n1 represents the real cost that the platformincurs for each type-2 developer, and a price p02 exactly equal to this real cost coincideswith the “socially optimal” price level of type-2 developers.

4 An Extension: Imperfect App Review

The previous section studies the case when the platform has perfect distinguishing abilitywhen reviewing apps, whereas in reality it may well make mistakes. If this happens, appreview can also reduce the utility of good developers. Suppose the app review processcan correctly identify all trouble-making developers, but with probability θ ∈ [0, 1], it willalso mistake a good developer for a trouble maker. θ can also be interpreted as the degreeof “imperfection” of app review.

Therefore under imperfect app review, when a type-2 developer obtains u2−R from theplatform, a type-1 developer expects to obtain a utility of u2 − θR (instead of u2 in section

10Note that the developers on Google and those on Apple may follow different accumulation functions(i.e. different φ2), and therefore their thresholds may also be different.

10

3). This changes how the number of type-1 developers under app review is determined,which now also depends on R and becomes

n21 = λφ2(u2 − θR).

Denote ∂n21∂R = −θλφ′2(u2 − θR). Nothing else changes in the previous model set-up.

The first order conditions for the optimal u1 and u2 are still given by (5) and (6), andthe platform’s incentive to use app review still depends on Π′(0) derived from (7), exceptthat because n21 also depends on R, we now have

Π′(R) =∂

∂Rπ(u∗1(R), u

∗2(R),R)

=∂n21∂R· (αn1 − u2 − f21) +

∂n22∂R· (αn1 − u2 − f22).

By (5) and (6), and let R = 0, we have

Π′(0) = (1 − θ)λ(1 − λ)∆ f · φ′2(u02) − (1 − λ + θλ) · φ2(u02),

which implies the following result.

Proposition 3 (Incentive to Review Apps with Imperfection) When app review is imper-fect, the platform has an incentive to introduce it if and only if

(1 − θ)λ(1 − λ)1 − λ + θλ ∆ f ≥

φ2(u02)

φ′2(u02)

(13)

where θ is the probability that app review mistakes a type-1 developer for type 2, and u02 is theutility that each developer obtains from the platform without app review, as given by (10).

Moreover, we can take the first order derivative of the left-hand side of (13) with re-spect to θ to find the impact of imperfection on the incentive to use app review. Because

∂ (1−λ)λ(1−θ)1−λ+θλ∂θ

=λ(λ − 1)

(1 − λ + θλ)2 < 0, (14)

we have the following result.

Proposition 4 The more imperfect the app review process is (i.e. the larger θ is), the less likelythat the platform will have an incentive to use it.

11

5 Welfare Analysis

5.1 General Framework

In order to clearly measure the welfare in the market, we now provide a specific way todefine the accumulation processes of users and developers that will result in the demandfunctions from two market sides described previously.

Assuming that a user derives an idiosyncratic value t1 if she participates on the plat-form, while her outside option of not participating gives her 0 utility. Therefore, giventhe utility that the platform provides to each user, u1, she participates if and only if

u1 + t1 ≥ 0.

Assuming the user population have heterogeneous idiosyncratic values that can bedescribed by distribution F1 on R, the accumulation process of users can be written as

φ1(u1) = Pr[u1 + t1 ≥ 0] = 1 − F1(−u1)

The function 1 − F1(−u1) is clearly strictly increasing in u1 and therefore is consistentwith our previous specification of φ1. This particular accumulation process of users allowsus to characterize their maximized total surplus as the following

v1(u1) ≡ Et1[max(u1 + t1, 0)] =∫ +∞−u1

(u1 + t)dF1(t)

which impliesv′1(u1) = φ1(u1)

Similarly, assume that a developer derives an idiosyncratic value t2 if she participateson the platform, while her outside option of not participating gives her 0 utility. There-fore, when the platform provides a utility of u2 to each developer, she participates if andonly if

u2 + t2 ≥ 0.

Assuming the developer population have heterogeneous idiosyncratic values that canbe described by distribution F2 on R, the accumulation process of developers can be writ-ten as

φ2(u2) = Pr[u2 + t2 ≥ 0] = 1 − F2(−u2)

The function 1 − F2(−u2) is also consistent with our previous specification of φ2.

12

Recall that a proportion of λ developers are good whereas the remainder are bad.When the platform uses app review, the utility it provides to bad developers is reducedby R compared to that provided to good developers, and therefore all developers’ totalmaximized surplus is dependent on both u2 and R, which we denote

v2(u2,R) ≡ λEt2[max(u2 + t2, 0)] + (1 − λ)Et2[max(u2 − R + t2, 0)]

= λ

∫ +∞−u2

(u2 + t)dF2(t) + (1 − λ)∫ +∞

R−u2(u2 − R + t)dF2(t)

which implies

∂

∂u2v2(u2,R) = λφ2(u2) + (1 − λ)φ2(u2 − R);

∂

∂Rv2(u2,R) = −(1 − λ)φ2(u2 − R)

The social welfare can therefore be represented by

w(u1, u2,R) = π(u1, u2,R) + v1(u1) + v2(u2,R). (15)

Our goal is to compare the different levels of social welfare market when the platformdoes and does not use app review, given that it maximizes profit in either case, usingthe optimal u∗1(R) and u

∗2(R) as characterized previously. That is, if we denote the social

welfare when the platform has maximized profit given R as

W(R) ≡ Π(R) + v1(u∗1(R)) + v2(u∗2(R),R)= π(u∗1(R), u

∗2(R),R) + v1(u

∗1(R)) + v2(u

∗2(R),R),

we will now compare W(R) and W(0) to determine whether app review will result in awelfare loss in the market.

13

The derivative of W(R) is

W ′(R) = Π′(R) + v′1(u∗1(R)) · u∗′1 (R)

+∂

∂u2v2(u∗2(R),R) · u∗′2 (R) +

∂

∂Rv2(u∗2(R),R)

= Π′(R) + φ1(u∗1(R)) · u∗′1 (R)+[λφ2(u∗2(R)) + (1 − λ)φ2(u∗2(R) − R)] · u∗′2 (R)−(1 − λ)φ2(u∗2(R) − R)

= Π′(R) + φ1(u∗1(R)) · u∗′1 (R) + λφ2(u∗2(R)) · u∗′2 (R)+(1 − λ)φ2(u∗2(R) − R) · (u∗′2 (R) − 1)

= Π′(R) + n1 · u∗′1 (R) + n2 · u∗′2 (R) − n22

An obvious point of interest is the sign of

W ′(0) = Π′(0) + n1 · u∗′1 (0) + n2 · u∗′2 (0) − n22 (16)

We will need to know u∗′1 (R) and u∗′2 (R), or at least some of their properties to tell the

sign of W ′(R) or W ′(0). As this is impossible without specifying the distributions of usersand developers (e.g. the F1 and F2 defined previously), we will use simulation.

5.2 Welfare Analysis under Uniform Distribution

Suppose the distributions of users and developers are uniform distributions. In particu-lar, users’ utility is distributed uniformly from a lower bound of a < 0 to an upper boundof b. App developers’ utility is distributed uniformly from a lower bound of d < 0 to anupper bound of c. Define B ≡ bb−a , and A ≡

1b−a (> 0). Define C ≡

cc−d , and D ≡

1c−d . Then we

can find the mass of users and app developers joining the platform offering u1 and u2 as:

φ1(u1) = B + Au1

φ2(u2) = C + Du2

14

So we have

n1 = B + Au1

n21 = λ(C + Du2)

n22 = (1 − λ)[C + D(u2 − R)]n2 = C + Du2 − (1 − λ)RD

And we also have

n11 = A, n111 = 0, n

221 = λD, n

222 = (1 − λ)D, n22 = D.

where the superscript j represents taking the partial derivative with respect to u j.The first-order conditions for the profit-maximizing u∗1 and u

∗2 are given by

(αn2 − u1 − f1) ·dn1du1

= n1

(αn1 − u2 − f22) ·dn2du2

= n2 − ∆ f ·dn21du2

Substituting the formulas under uniform distribution, we have

FOC(u1) : −2u1 + αDu2 = α(1 − λ)RD + f1 +BA− αC

FOC(u2) : αADu1 − 2Du2 = C + D f22 − (1 − λ)RD − ∆ f · λD − αBD

And the total differentiation of these equations with respect to R gives

dR of FOC(u1) : 2u∗′1 (R) − αD · u∗′2 (R) = −αD(1 − λ)dR of FOC(u2) : αA · u∗′1 (R) − 2u∗′2 (R) = −(1 − λ)

where the unique solution is given by

u∗′1 (R) =αD(1 − λ)α2AD − 4

u∗′2 (R) =(α2AD − 2)(1 − λ)

α2AD − 4

15

5.2.1 The special uniform distribution with A = D and B = C

In the special uniform distribution with A = D and B = C, i.e. if the distribution of twosides are the same, we have

u∗′1 (R) =αA(1 − λ)α2A2 − 4

u∗′2 (R) =(α2A2 − 2)(1 − λ)

α2A2 − 4

where we have the following property:

Claim i) If αA > 2, we have u∗′1 (R) > 0 and u∗′2 (R) > 0;

ii) if√

2 < αA < 2, we have u∗′1 (R) < 0 and u∗′2 (R) < 0; and

iii) if αA ≤√

2, we have u∗′1 (R) < 0 and u∗′2 (R) ≥ 0.

In this special case, we can also solve for the optimal u∗1 and u∗2 in FOC(u1) and FOC(u2)

shown previously, we are given by:

u∗1 =1

A3α2 − 4A(2B − ABα + 2A f1 + A2Rα − A2Rαλ + A2α f22 − A2αλ∆ f − A2Bα2

),

u∗2 =1

A3α2 − 4A(2B − 2AR − ABα + 2ARλ + 2A f22 − 2Aλ∆ f + A2α f1 − A2Bα2 + A3Rα2 − A3Rα2λ

).

The sign of W ′(0)

Now we can calculate everything in formula (16) in order to possibly sign W ′(0). Weare particularly interested in situations where Π′(0) > 0 but W ′(0) < 0, such that it wouldbe in the interest of the platform alone, not the society, to introduce app review. If this isthe case, the difference between W ′(0) and Π′(0) would also be negative.

A few simulated examples are provided next.

5.2.2 Simulation Results

We ran 503,119 times of simulation of the special case of uniform distribution discussedin the previous section. And the results show that the platform’s incentive to reviewapps, Π′(0), and the interest of social welfare, W ′(0), may be consistent in some casesbut divergent in others. Actually all four kinds of result exist, which are Π′(0) > 0 andW ′(0) > 0; Π′(0) > 0 and W ′(0) < 0; Π′(0) < 0 and W ′(0) > 0; Π′(0) < 0 and W ′(0) < 0.Though simulated, this is a very important result as it shows that anything is possible

16

even under rather simple uniform distributions. A regulator should therefore be verycareful when it contemplates whether to interfere with a platform’s app review process.Only in the situation when Π′(0) > 0 and W ′(0) < 0 can the intervention be justified.

The four sets of parameter values that yield the four different situations are providedas follows.

Example 1. Misalignment: Π′(0) > 0 and W ′(0) < 0 {A, B, α, f1, f22,∆ f , λ} = {0.5, 0.8, 2.4, 1, 2, 1.5, 0.1}

Π′(0) = 0.0288281

W ′(0) = −0.075542

Π(0) = 0.0473633.

Example 2. Alignment: Π′(0) > 0 and W ′(0) > 0 {A, B, α, f1, f22,∆ f , λ} = {0.5, 0.8, 1, 1, 2, 1.5, 0.1}

Π′(0) = 0.0915

W ′(0) = 0.0871

Π(0) = 0.0463333.

Example 3. Misalignment: Π′(0) < 0 and W ′(0) < 0 {A, B, α, f1, f22,∆ f , λ} = {0.5, 0.8, 3, 1, 2, 1.5, 0.1}

Π′(0) = −0.0353571

W ′(0) = −0.334745

Π(0) = 0.0564286.

Example 4. Misalignment: Π′(0) < 0 and W ′(0) > 0 {A, B, α, f1, f22,∆ f , λ} = {1, 0.9, 1.1, 2, 0.2, 0.15, 0.1}

Π′(0) = −0.0574677

W ′(0) = 0.0714316

Π(0) = 0.306837.

Furthermore, for the situation where Π′(0) > 0 and W ′(0) < 0, some simulation resultsalso satisfy Π(0) > 0, which means the platform is profitable under this situation and itwill conduct the review process, and the social welfare will decrease at the same time.There is only one situation that Π(0) > 0, Π′(0) > 0 and W ′(0) < 0 doesn’t exist at the sametime, that is when B = 0, which means that both developers and users have a uniformdistribution of negative utility derived from joining the platform. And in this situation,W ′(0) − Π′(0) is always bigger than 0, so introducing review process will always benefitsocial welfare first.

17

These results can be summarized as follows. More details are provided in the Ap-pendix.

Proposition 5 (Welfare) Simulation results show that the platform’s incentive to use app reviewmay be aligned or misaligned with the interest of social welfare, depending on the distributions ofapp users and developers and other parameters, including the strength of network effects α.

6 Conclusion

In this paper, we provide a rather straight-forward explanation for app review - it is in-tended to reduce the presence of trouble-making developers on the platform. When appusers have a preference for app variety despite the trouble some developers may imposeon the platform, the platform faces a trade-off between cost saving from screening troublemakers and a loss of the attractiveness to the users. We have shown that the platform’sincentive to use app review depends on the additional cost it incurs from each “troublemaker”, and on their proportion among all developers. Given the additional cost in-curred, app review benefits the platform if and only if the proportion of trouble makersis sufficiently small, such that after screening them the platform remains fairly attractiveto users. Given their proportion, however, app review helps if and only if the “troublemakers” create enough “trouble”, such that the total cost saved from app review is notnegligible. In other cases, the platform should not use app review, or should even use“reverse review” if feasible. This may explain the different app review strategies adoptedby Apple and Google.

Our welfare analysis through simulations cautions regulators that the platform’s in-centive to use app review may be aligned or misaligned with the interest of social welfare,depending on the distributions of app users and developers and other parameters.

Our model only considers a monopoly case. It can be applied to the smartphone mar-ket before the introduction of Google’s Android and Google Play, then known as theAndroid Marketplace, in late 2008. At that time, Apple was the dominant player in thesmartphone market. When smartphones were first introduced into the market, few peo-ple were able to develop mobile apps. Those app developers at that time were probablyIT professionals with exceptional programming skills and computer literacy. Their pref-erences, which can be measured by their time costs of developing apps, were less diversesince they usually had very high time costs. Thus it was profitable for Apple to introducea review process at that time.

The model may also explain why Google provides an open-source platform but Ap-

18

ple does not. The variety of smartphones using Android (where apps are distributed viaGoogle Play) is much higher than that using Apple’s iOS (where apps are distributed viaApp Store). The higher variety enables Google to attract app developers with a widerrange of preferences, whereas Apple may only attract a small range of app developers.For example, less capable app developers can develop simple apps for lower-priced andlower-quality Android smartphones, whereas more capable app developers can developcomplicated apps for higher-priced and higher-quality Android smartphones. Apple’sprofitable introduction of a review process therefore may not apply to Google. However,this analysis does not take into account the competition between the two platforms. Fur-ther research can analyze how the competition between the app distribution platformswill affect the platforms’ incentive to introduce a review process, as well as their pricingstrategies.

References

[1] Armstrong, Mark. 2006. “Competition in Two-sided Markets.” RAND Journal of Eco-nomics, 37(3): 668-691.

[2] Hagiu, Andrei. 2009. “Quantity vs. Quality and Exclusion by Two-sided Platforms.”Harvard Business School Working Paper, 09-094.

[3] Kouris, Iana. 2011. “Unified Two-sided Market Model.” Working paper.

[4] Rochet, Jean-Charles, and Jean Tirole. 2003. “Platform Competition in Two-sidedMarkets.” Journal of the European Economic Association, 1(4): 990-1029.

[5] Rochet, Jean-Charles, and Jean Tirole. 2006. “Two-sided Markets: A Progress Re-port.” Rand Journal of Economics, 37(3): 645-667.

Appendices

Appendix A Proofs and Derivations

Proof of Lemma 1

19

The partial derivative of (4) with respect to R gives

∂

∂Rπ(u1, u2,R) =

dn22dR

(αn1 − u2 − f21 − ∆ f )

wheredn22dR

= −(1 − λ)φ′2(u2 − R)

At the optimal utilities (u∗1(R), u∗2(R)), by (6) we have

(αn1 − u∗2(R) − f22) =n2 − ∆ f · dn21du2

∂n2∂u2

which implies

Π′(R) =∂

∂Rπ(u∗1(R), u

∗2(R),R) (17)

=dn22dR·

n2 − ∆ f · dn21du2∂n2∂u2

=−dn22dR

dn2du2

(∆ f · dn21du2− n2)

Therefore

∆ f · dn21du2− n2 = (

∂n2∂u2

−dn22dR) · Π′(R)

and

Π′′(R) =∂2

∂R2π(u∗1(R), u

∗2(R),R) (18)

=

dn21du2

(∂n2∂u2

)2· (−d

2n22dR2

) · (∆ f · dn21du2− n2) +

(dn22dR )2

∂n2∂u2

=

dn21du2

(∂n2∂u2

)2· (−d

2n22dR2

) · (∂n2∂u2

−dn22dR) · Π′(R) +

( dn22dR )2

∂n2∂u2

=

dn21du2∂n2∂u2

· (−d2n22dR2

) · ( 1−dn22dR

) · Π′(R) +( dn22dR )

2

∂n2∂u2

whered2n22dR2

= (1 − λ)φ′′2 (u2 − R)

Because ∂n2∂u2

> 0, and dn22dR , 0, we know Π′′(R) > 0 whenever Π′(R) = 0. �

20

Proof of Proposition 1

(9) and (10) are found by letting R = 0 in (5) and (6). �

Proof of Proposition 2

Let R = 0 in (17) and we have

Π′(0) = (1 − λ)[λ∆ fφ′2(u02) − φ2(u02)]

As φ′2(u02) > 0, we know

φ2(u02)φ′2(u

02)> 0. Therefore, whenever λ , 1 (i.e. there exist some type-2

developers), we have

Π′(0) > 0 if and only if λ∆ f >φ2(u02)

φ′2(u02).

Now consider the case when λ∆ f = φ2(u02)

φ′2(u02)

, that is, when Π′(0) = 0. By Lemma 1 weknow Π′′(0) > 0, therefore R = 0 is a minimizer of Π(R). Because introducing app reviewraises R from 0, it will make Π′(R) > 0 and is hence profitable. �

21

Appendix B Background Information on App Review

Apple has built a “walled garden” for its app store by pre-screening all apps and theirupdates before putting them onto its shelf.11 In this paper we will discuss 4 motivationsbehind Apple’s app review policy, including property protection, enhancement of brandloyalty, preservation of market power and upholding of the goodwill of the its own busi-ness. We will further highlight 2 catalysts that help to reinforce Apple’s determinationto install its app review policy. These are pickier consumers and lower training cost forstaff. We will go through relevant examples and data to support all the arguments made.

B.1 Introduction

A “walled garden,” or a closed platform, is a jargon used widely in the information tech-nology industry, describing a software platform introduced by a service provider or car-rier to control apps in respect of their content, media and applications, and to restricteasy access to ineligible content or applications.12 Roughly speaking, “walled garden”can be a phrase to describe some sort of isolated technology or platform, which is theopposite of an open source platform. Apple’s iOS is a typical example of a walled gar-den. Since Apple launched its exclusive Apple App Store in 2008, it has required appdevelopers to obtain its prior approval before putting their apps onto its shelf . All appshave to be pre-screened and Apple has full authority to reject or remove apps from itsapp store. Apple contends that its ex ante app review policy will ensure that the ap-proved apps are reliable, free of offensive content and at a level of high performance inorder to guarantee consumer satisfaction and to satisfy social responsibility.13 However,profit maximization, instead of consumers’ utility maximization, has long been the corepostulate of economics, making people query whether the reasons as advanced by Appleare really the true motives of Apple. Nevertheless, we cannot deny that maximizing theprofit of a firm may largely hinge on maximizing consumers’ utility. With this in mind,we seek to analyze in this paper the true motivations behind Apple’s app review policy.

11This part is the modified version of Douglas Cheng’s term paper entitled “The Walled Garden: Moti-vations behind Apple’s app review policy,” under the supervision of one of the authors.

12This definition comes from Wikipedia.13“We review all apps submitted to the App Store and Mac App Store to ensure they are reliable, perform

as expected, and are free of offensive material. As you plan and develop your app, make sure to use theseguidelines and resources,” extracted from https://developer.apple.com/app-store/review/

22

https://developer .apple.com/app-store/review/

B.2 Description of the app review

B.2.1 Background

Apple launched its Apple App Store in July, 2008, through an update to iTunes. As ofOctober, 2014, Apple App Store accumulated more than 85 billion downloads. Aroundthe same time, there were around 1.2 million apps available in Apple App Store, makingit the second largest app store in the world, behind Google Play which had 0.1 millionmore apps.14

B.2.2 App review guidelines

Apple has officially promulgated a set of strict guidelines, under the title of App StoreReview Guidelines, for its app review policy, under which both content and design of theapps will be examined.15 App developers who fail to comply with any of those guidelineswill have their apps rejected. I will not list the guidelines and describe them one by oneas it is not the interest of this paper. Instead, I summarize and generalize Apple’s focusduring its ex ante app review process on the basis of the App Store Review Guidelines.The next section notes some observations regarding the emphasis placed by Apple onupholding the review policy.

Basic guidelines can be largely generalized with reference to their functionality, con-tent appropriateness and legality. Apps with bugs, having non-functional hyper-links, orcrash will be rejected or removed from Apple App Store. Apps with inappropriate con-tent (insult to humanities, racism, pornography or personal attack) will be rejected. Also,the Apple review team will reject apps that infringe personal privacy, replicate other appsor fail to comply with local laws. It seems Apple is cautious of any sensitive or legal issuesthat will lead itself into trouble.

Apple puts much emphasis on user interface design as it wants to make sure that allthe apps are designed to suit its product but not the other way round.16 App developersneed to take care of even minor details when they create the user interface of their apps.Apple’s Design and Trademark Guidelines stipulate requirements for layout, interactiv-ity and integrality with iOS. For layout, app developers need to make sure that contentviews, temporary views, bars and controls of an app are placed at the right place withthe right size. For instance, the navigation bar and the tab bar have to be placed on thetop and bottom of the screen respectively, while the toolbar button has been placed at the

14http://www.statista.com/statistics/276623/number-of-apps-available-in-leading-app-stores/15The full list is at: https://developer.apple.com/app-store/review/16Supplementary guidelines for user interface design is at: https://developer.apple.com/design/

tips/

23

http://www.statista.com/statistics/276623/number-of-apps-available-in-leading-app-stores/https://developer .apple.com/app-store/review/https://developer.apple.com/design/tips/https://developer.apple.com/design/tips/

bottom right comer. Interactivity means the users of the app can see the primary con-tent without scrolling or zooming, and can comfortably make touch gestures. The textneeds to be at least 11 points and the hit targets need to be at least 44 points x 44 points.Integrality means that the app should incorporate as many general system-supportedtechnologies as it can, like VoiceOver and multitasking.

Also, the app should be able to integrate with some specific system-supported tech-nologies or functions where necessary, for example, Game Center, Passbook, iCloud, etc.By and large, it is preferable, if not necessary, for the app to integrate with iOS as muchas possible.

B.2.3 Common app rejections

It seems that Apple has a transparent and self-explanatory app review standard, with along list of guidelines clearly written in black and white. However, the mechanism maynot be as transparent as one imagines. In late-2014, Apple announced the top 10 commonreasons that apps were rejected.17

The most common reason, taking up 42% of the apps rejected app from App Storewas “Other reasons.” However, Apple did not further explain what exactly those otherreasons were. The second most common reason, taking up 14%, was equally obscure, forapps were rejected because of “more information needed.”

Some other common reasons of app rejections included non-functional or broken apps(8%), poorly designed user interface (6%) and irrelevant names, descriptions or screen-shots (5%).

B.2.4 Process of app review

As Apple has not officially disclosed how the app review mechanism works, so we canonly imagine the process with assistance of a disc10sure from an Apple’s employee.18

Mike Lee, a former senior engineer at Apple, suggested that Apple’s app review teamwas understaffed. He said, “like every other part of Apple, they can’t get enough reallygood people.” Lee also claimed “the app reviewers are essentially tasked with managinga slush pile of apps” and “there is so much garbage.” With a small app review team asclaimed, we can expect that each app reviewer has to deal with a multitude of apps eachday. When app reviewers reject apps in haste, it is not surprising that hidden gems maybe dumped accidentally as waste. Also, it takes around 8 days for Apple to review an

17The whole list is on https://developer.apple.com/app-store/review/rejections/18More information is at: http://www.businessinsider.com/heres-why-it-really-sucks-to-be-an-app-reviewer-for-apple-2012-7

24

https://developer.apple.com/app-store/review/rejections/http://www.businessinsider.com/heres-why-it-really-sucks-to-be-an-app-reviewer-for-apple-2012-7

app.19

B.3 Motives and examples

This section proposes Apple’s motivations behind its walled garden and lists the relevantexamples. There are 4 motivations behind Apple’s policy: property protection, enhance-ment of brand loyalty, preservation of market power and maintenance of the goodwill ofthe company.

B.3.1 Property protection

Apple is not a monopoly in the IT world, but it is the monopoly in the world of iOS.People who buy Apple products are actually buying a bundle of Apple hardware andiOS. Safari and iTunes are pre-installed, while Apple devices and iOS are exclusive toeach other. Apple is exceptionally mindful of protecting its technology, and app review isa way to combat property rights invasion. This idea is similar to other efforts and schemesto protect the “secret formula” of some franchises, like those adopted by Coca-Cola andPanadol.

Apple’s restrictions on app-discovery apps is a clear example to support the above ar-gument. An app discovery app is an application programming interface (API) that servesas a search engine for apps. It facilitates access to app database in different platforms andthen gathers, organizes and analyses the apps information for users. Examples of app-discovery apps include Mevvy API, App Shopper, Appidemia and App of the Day. Suchapps have not had a good fate in iOS as opposed to Android. According to MelanieHaselmayr, the CEO of Mevvy Limited, none of these apps were available before thelate-2013.20 All app-discovery apps were rejected, as Apple was too cautious to allow athird-party app to gain access to its database. In 2014, app-discovery apps started to ap-pear in the iOS market, because Apple started to provide those apps with its own set ofapp data. This means those app-discovery apps would no longer need to write an APIto extract information from Apple’s app database, and instead they would have limitedinformation provided by Apple. In comparison, there are a lot more app-discovery appson Google Play and these apps can have access to Android’s database.

19Information extracted from http://appreviewtimes.com20Information from Melanie Haselmayr comes from an email interview that 1 conducted in February,

2015.

25

http://appreviewtimes.com

B.3.2 Enhancement of brand loyalty

The fully integrated iOS ecosystem is a phenomenal feature of Apple Inc., i.e., all iProd-ucts are branded in harmony together with Apple’s exclusive operating system. All Ap-ple products look similar, so are their functionality. For instance, iPad is an enlarged iPodwhere iPhone is an iPod with telecommunication functions. Apple wants to ensure thatan app is functional on one kind of device and is equally functional on another productof the company. This explains a reason for using the app review policy to eliminate appsthat fail to observe the design user interface standard or are not compatible with generalsystem-supported technology, as in the case of Siri. Thus, there will be easier and tighterinter-operability with the OS. As a result, Apple will increase consumers’ switching costas users are too used to Apple products as well as iOS. Apple is achieving a 76% customerretention rate which is the highest in the mobile industry.21 On the contrary, if the appdesign is too random, or too gadget-specific, consumers will be more prone to switchingto another brand or another platform operated by a rival company.

B.3.3 Preservation of market power

The app review policy can be anti-competitive as it can be a gimmick to drive away poten-tial competitors. In particular, there may be a business field into which Apple is desirousof making inroads. It may therefore take ear1y steps to reject or remove competitors’apps in that field. Bitcoin and Google Voice are two landmark examples as explained inthe following paragraphs.

In early-2014, Apple started removing Bitcoin apps without any explanation. Thiswas a controversial move.22 Bitcoin is a virtual currency that people can use to pay forgoods and services and it is becoming popular worldwide. Many suspected that Applewithdrew Bitcoin apps from App Store because Apple was developing its own way ofvirtual payment, i.e. the Apple Pay.23 Bitcoin could be a big threat Apple Pay. But laterin the same year, Bitcoin apps were available on iOS again. It was because after Appleremoved all Bitcoin apps, Google became the only platform allowing Bitcoin apps. Appledid not want to want to lose out to Google, and had therefore relented in the case ofBitcoin. It is all about money.

Another example is rejection of Google Voice from App Store in 2009. Apple even pub-

21Information extracted from: http://news.xerox.com/news/WDS-A-Xerox-Company-Mobile-Loyalty-Audit-Results-201422For more information see http://www.coindesk.com/apple-removes-blockchain-bitcoin-wallet-from-app-stores/23For more information see: http://www.pcworld.com/article/2095060/

apple-removes-blockchain-last-bitcoin-wallet-app-from-mobile-store.html

26

http://news.xerox.com/news/WDS-A-Xerox-Company-Mobile-Loyalty-Audit-Results-2014http://www.coindesk.com/apple-removes-blockchain-bitcoin-wallet-from-app-stores/http://www.pcworld.com/article/2095060/apple-removes-blockchain-last-bitcoin-wallet-app-from-mobile-store.htmlhttp://www.pcworld.com/article/2095060/apple-removes-blockchain-last-bitcoin-wallet-app-from-mobile-store.html

licly gave an explanation.24 Following up on Apple’s explanation, Google Voice sought toreplace a lot of core functions of iPhone, but Apple did not appreciate that. Apple is leastkeen to enable its users to experience another user interface or core functions other thaniOS. Now that Google Voice is available on iOS because the design has been modified tosuit the iOS system. Therefore, app review is a powerful and potentially anti-competitivepractice.

B.3.4 Maintenance of the goodwill of the business

Apple puts more emphasis on end users experience than Google because hardware salesare its mainstream revenue.25 Hardware sales hinge much on the goodwill of the businesswhich can be improved through better customer experience. If there are some sensitiveapps that maybe regarded as offensive to religion or humanities, the offended individualsmay boycott Apple’s products, leading to a big loss of sales revenue for the company. Thatis why Apple is very careful when it comes to eliminating troublemakers from App Storein order to sustain and not to tarnish its reputation. Generally, apps with obscene, racist,gambling or violence content will not be allowed.

There are many more relevant examples regarding sensitive content being rejected.Here, I am going to briefly discuss two.

In 2011, “The Unpleasant Horse” was rejected from iOS because the app was seen tobe promoting violence against animals.26 This app is a game that when a player reachesthe stage of ”Game Over”, the horse will be grinded into mince meat. Apple did notallow such an app to exist in its app store to avoid complaints from animal lovers. An-other example is “Baby Shaker,” a game where players shake the crying animated babyto death.27 Apple initially approved this app but it then caused a lot of controversy as thegame was seen to be promoting infanticides. Many parents expressed their discontentby suggesting that they would boycott Apple products. Apple then quickly removed theapp with an apology. This illustrates that problematic apps can harm a company’s image.

24The whole statement is at: https://www.apple.com/hotnews/apple-answers-fcc-questions/25In Q4 of 2014, around 86% of Apple’s revenue came from hard-

ware sales. Information extracted from http://www.macstories.net/news/apple-q4-2014-results-42-1-billion-revenue-39-3-million-iphones-12-3-million-ipads-sold/

26More information on http://www.digitaltrends.com/mobile/apple-rejects-popcaps-unpleasant-horse-due-to-violence/27More information at:http://www.theguardian.com/technology/2009/apr/23/

apple-iphone-baby-shaker

27

https://www.apple.com/hotnews/apple-answers-fcc-questions/http://www.macstories.net/news/apple-q4-2014-results-42-1-billion-revenue-39-3-million-iphones-12-3-million-ipads-sold/http://www.macstories.net/news/apple-q4-2014-results-42-1-billion-revenue-39-3-million-iphones-12-3-million-ipads-sold/http://www.digitaltrends.com/mobile/apple-rejects-popcaps-unpleasant- horse-due-to-violence/http://www.theguardian.com/technology/2009/apr/23/apple-iphone-baby-shakerhttp://www.theguardian.com/technology/2009/apr/23/apple-iphone-baby-shaker

Table 1: Comparing iOS and Android apps

iOS AndroidApps count 1,384,043 976,444Mean user rating 3.6337 3.9918Mean rating count 112.40 1,591.34Percentage of free apps 71% 85%Number of common apps 221,321Mean price for common apps $0.6886 $0.34002

B.4 Data and statistics

This section examines Apple’s app review policy against some data. App data from iOSand Android were given by Mevvy Ltd on 1st March, 2015. Below are some statistics:

We have almost all app data pertaining to iOS, while only a subset of data for Android.As of March 2015, Android had around 0.1 million more apps than iOS. Sti11, the closenumber of apps between the two platforms is surprising, probably due to several reason.Firstly, Android apps are easier to develop in programming sense, and app developersmostly agree with this. Secondly, trial versions, beta versions, light versions are allowedon Android but not on iOS. Thirdly, the world market share of Apple products is onlyone-third of Android devices. Fourth, and perhaps most importantly, Android has no pre-screening of apps. Therefore, for iOS to have nearly as many apps available as Android isquite unexpected, and such phenomenon is worthy of another specific study to examinethe causes and to ascertain answers.

Mean user rating is a 5-star scale rating which measures the average consumer sat-isfaction. From our sample, the mean user rating for iOS was 0.36 lower than that ofAndroid. This may imply that the iOS users were more difficult to serve, or in otherwords they were pickier. Mean rating count shows how many users rate the app on av-erage, acting as a proxy for the number of downloads per app. Since iOS did not publiclydisclose to or provide Mevvy Limited with the data for downloads per app, we can onlyuse a proxy. It is not surprising to see Android had much higher mean rating per app asit had a higher market share worldwide. Furthermore, there were 221,321 common appsacross the 2 platforms, while the mean price for those common apps on iOS was almostdouble that of Android.

Figure ?? graphs that plot price range against average rating count. For the sake ofeasier data handling, 1 have grouped the apps into 6 price ranges: P = $0; $0 < P ≤ $5;$5 < P ≤ $10; $10 < P ≤ $15; $15 < P ≤ $20, and P > $20.

From the above graphs, we can see there is a steeper slope for the “demand curve” for

28

iOS. This may imply that Apple users were more willing to pay for apps than Androidusers. It should be noted that both x- and y-axis are only proxies for P and Q of a normaldemand curve. Thus this implication is not definite.

From all above graphs and numbers, we can derive one implication. Apple users andAndroid users do not seem to share common characteristics and consumer behavior. Theformer group comprises those more willing to pay for apps, and more demanding. Thismeans that problematic apps will trigger more problems for iOS than Android. This maybe a justification, or some may say a driving force, for Apple to implement an ex ante appreview policy.

B.5 Counter-example: Google’s Android

Sometimes when we think ”Why A?”, it may be equally if not more inspiring to think”Why not B?”. In this context, “Why not Google?” The previous sections have explored acouple of motivations behind Apple’s app review policy, where such a policy can protectApple’s business. One may ask why Google does not follow suit and practice the samepre-screening policy.

I propose 2 main reasons: Google’s main source of revenue is not from hardware salesand the fixed cost incurred to train and operate a review team is too high.

To further elaborate on the main reasons, let us briefly examine the main differencesbetween Apple’s and Google’s business models. Apple has a couple of roles in its busi-ness model, where it started as a computing hardware developer. Subsequently, it openedthe iPhone line with a unique operating system, thus becoming a mobile device manu-facturer as well as a computing software developer. Also, Apple is an e-commerce engineand retailer. The business roles of Apple are well vertically integrated and more versatilethan Google which is mainly an internet service provider.

Unlike Apple, Google’s main source of revenue comes from advertising (around 40%),where its hardware sales only take up a slight portion of its earnings (less than 10%).28

Android is not exclusive to Google’s devices as there are more gadget producers likeSamsung, Sony, LG that also support the mobile operating system. Owning only Mo-torola, Google is far from a dominant competitor in the market of mobile devices. Theopportunity cost of selling some potentially troublesome apps will be lower for Googlethan Apple. Therefore, Google is less concerned about losing sales owing to some sensi-tive apps, while the company is more inclined to allowing more apps on Google Play togenerate advertising revenue through Google Mobile Ads.

28As of Dec 31, 2013, hardware sales only take up 7% of Google’s revenue for a quarter. More informationat: https://investor.google.com/financial/tables.html

29

https://investor.google.com/financial/tables.html

Secondly, 1 am not saying that Google would never remove problematic apps from itsapp store. Indeed, Google tends to do that as well. If some users discover and report thatsome apps are scam, broken, or likely to cause legal trouble, Google will seek to removethem. For example, in 2014, there was a fake anti-virus app called Virus Shield whichonce topped the Google Play’s sales chart.29 Google subsequently removed the app andeven provided refunds to its customers. This illustrates that Google does care about thegoodwill of its brand name but it is also undeniable that Google will incur huge fixedcost for training an app review team for the sake of pre-screening apps. As mentioned inthe previous paragraph, there are a wide variety of mobile devices that support Androidwhile the majority of them are not owned by Google. An app works well on one gadgetis no guarantee that it works or looks equally well on another gadget. This means ifGoogle seeks to practice an ex ante app review policy, there will be huge fixed trainingcost incurred. So it is better for Google to remove problematic apps as per customers’requests. In comparison, App Store is exclusive to Apple products, where Apple knowsits devices well, enabling it to spend much lower fixed cost than Google to train a teamof app reviewers. Moreover, as mentioned above, Apple has the role of a retailer in itsbusiness model. As a retailer, Apple has a front line that constantly keeps in touch withcustomers, enabling mutual communication and collection of customer opinions. As aresult, Apple has a better chance of making the right decision when it comes to approvingapps for inclusion in the Apple App Store.

B.6 Conclusion

To sum up, we have gone through 4 motivations, i.e., property protection, enhancementof brand loyalty, preservation of market power and maintenance of the goodwill of thebusiness, and a further 2 driving forces, i.e., pickier consumers and lower training costfor staff in implementing Apple’s app review policy. There are fundamental differencesbetween Apple’s and Google’s business models that make only the former practices apre-screening policy.

Steve Job once said, “We do believe we have a moral responsibility to keep porn offthe iPhone. Folks who want porn can buy an Android phone.” In a similar line, the Appleofficial website says “People who want to criticize a religion and culture can choose writea book.” Apple may profess to want to fulfill social responsibility, but the prime incentivefor a profit-making firm has always been profit-making. If it is more profitable and lessrisky to carry out an app review policy, a company will have incentives to implement one.

29More information at: https://nakedsecurity.sophos.com/2014/04/22/google-refunds-android-users-who-bought-fake-virus-shield-app/

30

https://nakedsecurity.sophos.com/2014/04/22/google-refunds-android-users-who-bought-fake-virus-shield-app/https://nakedsecurity.sophos.com/2014/04/22/google-refunds-android-users-who-bought-fake-virus-shield-app/

B.7 References

[1] Arthur, C. (2009, April23). “ ‘Shaker’ game pulled from Apple’s iPhone App Store.”Retrieved from http://www.theguardian.com/technology/2009/apr/23/apple-iphone-baby-shaker

[2] Fiegerman, S. (2012, July 3). “Here’s Why It Really Sucks To Be An App ReviewerFor Apple.” Retrieved from http://www.businessinsider.com/heres-why-it-really-sucks-to-be-an-app-reviewer-for-apple-2012-7

[3] Kirk, J. (2014, February 6). “Apple removes Blockchain, last Bitcoin wallet app,from iOS App Store.” Retrieved from http://www.pcworld.com/article/2095060/apple-removes-blockchain-last-bitcoin-wallet-app-from-mobile-store.html

[4] Munson, L. (2014, April22). “Google refunds Android users who bought fakeVirus Shield app.” Retrieved from https://nakedsecurity.sophos.com/2014/04/22/google-refunds-android-users-who-bought-fake-virus-shield-app/

[5] Southurst, J. (2014, February6). “Apple Removes Blockchain Bitcoin Wallet Appsfrom its App Stores.” Retrieved from http://www.coindesk.com/apple-removes-blockchain-bitcoin-wallet-from-app-stores/

[6] Spencer, G. (2014, October 21). “Apple Q4 2014 Results: $42.1 Billion Revenue, 39.3Million iPhones, 12.3 Million iPads Sold.” Retrieved from https://www.macstories.net/news/apple-q4-2014-results-42-1-billion-revenue-39-3-million-iphones-12-3-million-ipads-sold/

[7] Van Camp, J. (2011, April 8). “Apple rejects PopCap’s Unpleasant Horse due to ex-cessive pony meat grinding.” Retrieved from http://www.digitaltrends.com/apple/apple-rejects-popcaps-unpleasant-horse-due-to-violence/

[8] Xerox. (2014, February 24). “Apple and Samsung vs the rest.” Retrieved fromhttp://www.wds.co/apple-samsung-vs-rest/

31

http://www.theguardian.com/technology/2009/apr/23/apple-iphone-baby-shakerhttp://www.businessinsider.com/heres-why-it-really-sucks-to-be-an-app-reviewer-for-apple-2012-7http://www.pcworld.com/article/2095060/apple-removes-blockchain-last-bitcoin-wallet-app-from-mobile-store.htmlhttp://www.pcworld.com/article/2095060/apple-removes-blockchain-last-bitcoin-wallet-app-from-mobile-store.htmlhttps://nakedsecurity.sophos.com/2014/04/22/google-refunds-android-users-who-bought-fake-virus-shield-app/https://nakedsecurity.sophos.com/2014/04/22/google-refunds-android-users-who-bought-fake-virus-shield-app/http://www.coindesk.com/apple-removes-blockchain-bitcoin-wallet-from-app-stores/https://www.macstories.net/news/apple-q4-2014-results-42-1-billion-revenue-39-3-million-iphones-12-3-million-ipads-sold/https://www.macstories.net/news/apple-q4-2014-results-42-1-billion-revenue-39-3-million-iphones-12-3-million-ipads-sold/http://www.digitaltrends.com/apple/apple-rejects-popcaps-unpleasant-horse-due-to-violence/http://www.digitaltrends.com/apple/apple-rejects-popcaps-unpleasant-horse-due-to-violence/http://www.wds.co/apple-samsung-vs-rest/

IntroductionThe modelWhen to Review Apps?An Extension: Imperfect App ReviewWelfare AnalysisGeneral FrameworkWelfare Analysis under Uniform DistributionThe special uniform distribution with A=D and B=CSimulation Results

ConclusionAppendicesAppendix Proofs and DerivationsAppendix Background Information on App ReviewIntroductionDescription of the app reviewBackgroundApp review guidelinesCommon app rejectionsProcess of app review

Motives and examplesProperty protectionEnhancement of brand loyaltyPreservation of market powerMaintenance of the goodwill of the business

Data and statisticsCounter-example: Google's AndroidConclusionReferences